FIX parse likert raw score via float before int

[WatchTree-19]

problem

SelfAskLikertScorer raises ValueError when the judge returns a numeric likert value with a decimal point (e.g. "4.0"), which models commonly do for a "numeric value between 1 and 5".

in _score_async the raw value is parsed two different ways:

• score value (works): float(unvalidated_score.raw_score_value)
• metadata (crashes): int(unvalidated_score.raw_score_value)

raw_score_value is str(parsed_response["score_value"]), so a judge response of 4.0 becomes "4.0". float("4.0") is fine, but int("4.0") raises ValueError: invalid literal for int() with base 10: '4.0', so the whole score fails even though the float-scaled value computed correctly. SelfAskScaleScorer doesn't hit this because it never int()s the raw value.

fix

parse the metadata via float first, matching the score-value line:

score.score_metadata = {"likert_value": int(float(unvalidated_score.raw_score_value))}

test

added test_likert_scorer_accepts_float_string_score_value (judge returns score_value: "4.0"). fails on main (ValueError), passes with the fix, and asserts the likert_value metadata is 4. full tests/unit/score/test_self_ask_likert.py 34 pass; ruff check + ruff format clean.

found by reading the scorer, not from the tracker. happy to file an issue first if you'd prefer.